Large deviations for the hydrodynamical limit of independent asymmetric random walks (Q1346403): Difference between revisions

The authors prove large deviation principles for the hydrodynamical limit of the measure-valued process \[ Y^ N_ t = N^{-d} \sum_{i \in \mathbb{Z}^ d_ N} \eta_ t (i) \delta_{i/N}, \quad t \in [0,T],\;N \in \mathbb{N}, \] where \(\mathbb{Z}^ d_ N = (\mathbb{Z}/N \mathbb{Z})^ d\) and \(\eta_ t (\cdot)\) is an \(\mathbb{N}^{\mathbb{Z}^ d_ N}\)-valued Markov process which corresponds to a superposition of (asymmetric) independent random walks jumping at \(i \in \mathbb{Z}^ d_ N\) with intensity \(c(i) \geq 0\). If the deterministic initial conditions \(Y^ N_ 0 \equiv \rho^ N_ 0\) converge weakly (in the sense of measures) to a measure \(\rho_ 0\), then the following upper bound \[ \limsup_{N \to \infty} N^{-(d+1)} \log P(Y^ N \in C) \leq - \inf_{\mu \in C} I(\mu) \] holds for all closed subsets \(C\) in the Skorokhod space \(D([0,T], {\mathcal M}^ +)\), \({\mathcal M}^ +\) being the bounded positive measures on \((\mathbb{R}/ \mathbb{Z})^ d\). If \(d = 1\) and if \(\eta_ t\) is nondegenerate, i.e. has jumps in both directions, a corresponding minoration formula is given. In both cases, the rate function \(I(\mu)\) is calculated in a nonvariational form. It turns out, that \(I(\mu)\) is the conjugate function (Legendre transform) of a -- due to the asymmetry -- nonquadratic function \(\tau\). The rôle of \(L^ 2\) of the symmetric case is taken over by Orlicz spaces associated with \(\tau\). The methods employed here, rely heavily on techniques from convex analysis.